Releases·Tool Release
AIPass ships v2.7.20 and v2.7.21, rebuilding agent memory and repo-shipped passports
AIPass released two versions, v2.7.20 and v2.7.21, in its Update #17 dev log. The releases rebuild the two components the author identifies as making an agent an agent: its memory and its passport. The update series is now folded into the project's QA loop.
4 engagement1 sourceSun, Aug 30, 2026, 04:17 AM
AI Overview
- Two releases shipped since Update #16: v2.7.20 and v2.7.21, the second tagged the night of the post
- Memory and passport — described as the two files that make an agent an agent — were both 'torn down to the studs and rebuilt'
- Passports now ship with the repo per the release
- The missing v2.7.17 changelog header flagged in Update #16 was fixed the same night, with release notes crediting the find to the update series
- The update series is now part of the QA loop
Entities
Related
Social·Community
AIPass memory design: agents own small JSON files instead of vector stores
A developer's deep dive into the AIPass repo describes a multi-agent memory system where each agent ('citizen') owns a .trinity directory with three plain JSON files, replacing large vector stores. The design emphasizes small, session-limited files for identity, history, and observations, aiming for efficiency and clarity in multi-agent setups.
A developer's deep dive into the AIPass repo describes a multi-agent memory system where each agent ('citizen') owns a .trinity directory with three plain JSON files, replacing large vector stores. The design emphasizes small, session-limited files for identity, history, and observations, aiming for efficiency and clarity in multi-agent setups.
2 engagement1 sourceredditSun, Aug 16, 2026, 07:22 AM
Products·building
Heimdall: Self-healing graph memory for coding agents
Heimdall is a graph-based knowledge system for coding agents that maintains semantic retrieval with probabilistic strength ratings. It self-heals by watching commands and inferring edits, eliminating stale data and bash loops. It solves the problem of agents getting stuck or colliding over files across multiple projects and repositories.
Heimdall is a graph-based knowledge system for coding agents that maintains semantic retrieval with probabilistic strength ratings. It self-heals by watching commands and inferring edits, eliminating stale data and bash loops. It solves the problem of agents getting stuck or colliding over files across multiple projects and repositories.
45 engagement1 sourceredditSat, Aug 22, 2026, 02:32 AM
Releases·Tool Release
Brain-AI Memory v0.4.0 released: local memory kernel for long-running agents
A developer released Brain-AI Memory v0.4.0, an MIT-licensed, installable, local, provider-neutral reference kernel that separates episodic events, semantic knowledge, procedural rules, exact state, entities/relations, and lifecycle decisions. The post highlights that long-running agents can retrieve the right text and still fail due to stale decisions, mixed projects, or abandoned fallback sequences, and that treating these as retrieval problems makes diagnosis difficult.
A developer released Brain-AI Memory v0.4.0, an MIT-licensed, installable, local, provider-neutral reference kernel that separates episodic events, semantic knowledge, procedural rules, exact state, entities/relations, and lifecycle decisions. The post highlights that long-running agents can retrieve the right text and still fail due to stale decisions, mixed projects, or abandoned fallback sequences, and that treating these as retrieval problems makes diagnosis difficult.
1 engagement1 sourceredditWed, Jul 15, 2026, 09:05 AM
Products·launched
Claude memory staleness detector and fixer
A tool that detects stale or misleading entries in Claude's memory files, flagging outdated claims and closed incidents that could mislead future sessions. It helps developers maintain accurate agent memory by identifying and correcting obsolete information.
A tool that detects stale or misleading entries in Claude's memory files, flagging outdated claims and closed incidents that could mislead future sessions. It helps developers maintain accurate agent memory by identifying and correcting obsolete information.
9 engagement1 sourceredditTue, Aug 11, 2026, 06:34 AM
Products·building
Persistent multi-agent family using identity files, file-based message bus, and forward-written journals
A non-engineer built a system of 13 named AI assistants that operate as one continuous operation across different apps, using identity files for quick bootstrapping, a file-based message bus for reliable communication, and forward-written journals for session continuity. It solves the problem of session death in LLM interactions, enabling persistent, collaborative AI workflows without a server.
A non-engineer built a system of 13 named AI assistants that operate as one continuous operation across different apps, using identity files for quick bootstrapping, a file-based message bus for reliable communication, and forward-written journals for session continuity. It solves the problem of session death in LLM interactions, enabling persistent, collaborative AI workflows without a server.
29 engagement1 sourceredditWed, Aug 19, 2026, 06:33 AM